A pixel-specific model observer for high-fidelity assessment of x-ray angiography imaging system performance

Objective. While clinical tasks require visualization of spatially complex interventional devices, typical methods to assess X-ray angiography systems performance utilize stylized test objects. The purpose of this work was to develop methods to estimate the contribution of individual X-ray angiography system pixels to test object SNR 2 , thus providing for the use of clinical devices as test objects. Approach. Theory to estimate the contribution of individual pixels (snt 2 ) to test object SNR 2 was derived. For each pixel, the ‘signal’ was the mean pixel value difference between a set of test signal present and absent images. ‘Noise’ was estimated as the pixel-specific total covariance (ω 2 ) calculated within a local pixel neighborhood. Test object SNR from this pixel-specific observer (PSO) was compared to that from a Hotelling observer which used pixel values as variables (PHO) and a Gabor channelized Hotelling observer (CHO). Bias from finite sampling and detector electronic noise was managed by established methods. Image samples were acquired using stationary iodine disks (diameter 0.5 to 4.0 mm) and pediatric interventional devices as test objects. Quantum fluence was varied over a range of 8x to mimic exposure conditions for both fluoroscopy and angiography. Main results. For the iodine disk test objects, SNR was in the range ~1.0 to 32. SNR estimates by the PSO method were linear with object size, X-ray fluence, and PHO and CHO estimates. Pixel-specific snr 2 maps of implantable devices demonstrated that the pixel-specific approach preserved the high spatial fidelity of the X-ray angiography system, including the effects of focal spot blur. Significance. That test object SNR estimates by the pixel-specific methods were consistent with those from the PHO and CHO provides mutual support for the accuracy of each method. The newly proposed approach provides for the use of spatially complex clinical devices as test objects.